The correct answer is C. Cloud Spanner distributes data based on primary keys. Using Version 1 UUIDs, which include timestamps and increase monotonically, causes new rows to be written to the same split (hotspotting). This prevents linear scaling as writes are concentrated on a single node, even as the cluster grows. Options A and B relate to data type inefficiencies but do not directly impact scaling. Option D affects query performance but not the underlying distribution of data across nodes.